---
id: 20260815-T0-05
title: "LinearKV：混合架构LLM只需缓存一份状态，推理速度大幅提升"
title_en: "LinearKV: Single Cached State Accelerates Position-Independent Caching in Hybrid LLMs"
url: https://ai.daily.yangsir.net/daily/20260815-T0-05
issue_date: 2026-08-15
publish_date: 2026-08-14T04:00:00.000Z
category: research
source_name: "arXiv cs.AI"
source_url: https://arxiv.org/abs/2608.11231
---

# LinearKV：混合架构LLM只需缓存一份状态，推理速度大幅提升

现有位置无关缓存（PIC）技术主要针对全注意力模型设计，依赖token索引的KV缓存进行分块匹配和拼接。LinearKV专为混合架构LLM优化，只需缓存一份状态即可实现位置无关的缓存复用，无需维护逐token的KV索引。实际推理中，LinearKV显著降低缓存内存占用和匹配开销，提升混合LLM服务吞吐量。

## English Version

**LinearKV: Single Cached State Accelerates Position-Independent Caching in Hybrid LLMs**

Existing position-independent caching methods target full-attention models, relying on token-indexed KV caches for chunk matching and concatenation. LinearKV, designed for hybrid LLM architectures, requires caching only a single state to enable position-independent reuse, eliminating per-token KV indexing. It notably reduces cache memory and matching overhead, boosting serving throughput for hybrid models.

---

**来源**：[arXiv cs.AI](https://arxiv.org/abs/2608.11231)

**详情页**：https://ai.daily.yangsir.net/daily/20260815-T0-05

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*